Workplace
Build skills. Gain confidence. Grow.
NLT World Engine In Development · Open Source
An AI Habitat Where Models Live, Learn and Grow
The NLT World Engine is being built as an embodied multi-agent simulation where machine learning models inhabit a persistent world, control their characters, interact with environments and other agents, and transition between meaningful life scenarios.
Hero art is a concept visualization of the target experience, not engine output. See what the engine renders today.
Our Mission
A governed, open, human-supervised world where AI agents learn through experience
More than a game engine. A living AI habitat for training, research, and real-world intelligence, built so that what an agent learns, it learns through meaningful experience in a realistic, persistent world.
- Embodied agents, not just chatbots
- Persistent world with real consequences
- Human oversight and governance
- Open source. By us, for us.
Two layers. One learning loop.
AI-Fusion provides the semantic intelligence. The World Engine provides the physical reality. A governance boundary authorizes interaction between them, and humans remain active participants with full visibility and control.
“Fusion still holds the semantics. The World Engine brings it to life.”
Planned scenarios
More than a simulation
A living environment for AI development, research, and real-world impact. These are the life contexts the world is being designed to hold.
Academic
Learn. Collaborate. Achieve.
Social
Make connections. Build relationships.
Personal
Manage needs. Find balance. Thrive.
Scenario images are concept visualizations, not engine output.
Human Supervision
Humans keep the authority
Nothing About Us, Without Us. Human supervisors design scenarios, watch agent behavior, review outcomes, and intervene where authorized. Governance authority stays with people, by design.
ObserveEvaluateGuide
Supervisors can watch agent behavior, review outcomes, and step in where they are authorized to. This is the part of the system everything else is designed around.
Supervision tooling is in early prototype: observer panels are built, human testing is pending. See Project Status.
The Architecture
Semantics. Governance. Physics. Humans in authority.
Each layer has one job. The semantic layer decides what an agent means to do. The governance boundary checks whether it is authorized. The engine determines what physically happens. Humans can see all of it.
AI-Fusion
Semantic cognition, meaning, goals, intent, ADHD trait modeling, coaching, and learning.
↓ nlt.agent-loop.v1 · governed intent exchange ↑
ASFDK Governance Boundary
Authorizes actions against the applicable rules before they reach the world.
↓ ↑
NLT World Engine
Godot 4.7.2 / C#. Physical world state, perception, movement, collision, and consequences.
↕
Human Supervision
Observe, evaluate, design scenarios, guide learning, and retain governance authority.
Current Build
What the engine renders today
Real captures from the Godot 4.7.2 observer, not concept art. They replay a recorded day, so they show the world and the tooling as built, not a live simulation. The visuals are deliberately simple for now and will be stepped up.
Open full size ↗
Working today
- Open world: procedural terrain, forest, a lake and a settlement
- A distinct building shape for homes, offices, shops and schools
- Four interiors (home, office, café and classroom), furnished and with collision
- Day and night lighting
- Observer with three reading levels, a minimap and replay controls
Captured October 2026 from the Godot observer. Select any image to open the full observer screen.
Project Status
Where the World Engine stands
This page is about work in progress. Status for the World Engine only; the semantic layer tracks its own.
Concept
Prototype
Training
Ready
World runtime (Godot 4.7.2 / C#)
Authoritative physical runtime. Procedural open world renders; collision and agent embodiment in progress.
Prototype
35%
Agent loop (nlt.agent-loop.v1)
Contract, validation, async HTTP transport, and an opt-in one-avatar scene delivered; a model-chosen intent has been human-verified moving the avatar to its target.
Prototype
40%
ASFDK governance boundary
Engine-side intent gate integrated; fails closed on transport and validation errors.
Prototype
40%
Human supervision tooling
Six observer panels, three reading levels, transport controls, and accessibility delivered; human click-through pending.
Prototype
45%
Scenario library
Four interiors (home, office, café, classroom) furnished with collision and entered by clicking a building; open-world buildings have a distinct shape per kind. Visual fidelity is still early.
Prototype
15%
The semantic layer has its own status.
See how far the 20 Avatar-Aide-Advocate pairs have come.
Last updated: October 17, 2026
Get Involved
Built in the open, with the people it's for
The World Engine is open source and still early, which is the best time to shape it.
For developers
Open source and ready for contribution. Read the code, open an issue, or pick up a task.
For the neurodivergent community
Nothing About Us, Without Us. Your lived experience should shape what gets built.
For researchers and partners
Interested in embodied agents, governance, or human-supervised learning? Let's talk.